4

I have two vectors of floats, x and y, and I want to compute the Pearson correlation coefficients. As I have to do it on a lot of data (for instance 10 millions different vectors x and 20 thousand different vectors y), I am using C++, and more specifically the gsl_stats_correlation function from the GSL.

Here is my C++ code:

#include <iostream>
#include <vector>
using namespace std;

#include <gsl/gsl_vector.h>
#include <gsl/gsl_statistics.h>

int main (int argc, char ** argv)
{
  vector<double> x, y;
  size_t n = 5;
  x.push_back(1.0); y.push_back(1.0);
  x.push_back(3.1); y.push_back(3.2);
  x.push_back(2.0); y.push_back(1.9);
  x.push_back(5.0); y.push_back(4.9);
  x.push_back(2.0); y.push_back(2.1);
  for(size_t i=0; i<n; ++i)
      printf ("x[%ld]=%.1f y[%ld]=%.1f\n", i, x[i], i, y[i]);

  gsl_vector_const_view gsl_x = gsl_vector_const_view_array( &x[0], x.size() );
  gsl_vector_const_view gsl_y = gsl_vector_const_view_array( &y[0], y.size() );

  double pearson = gsl_stats_correlation( (double*) gsl_x.vector.data, sizeof(double),
                                          (double*) gsl_y.vector.data, sizeof(double),
                                          n );
  printf ("Pearson correlation = %f\n", pearson);

  return 0;
}

It compiles successfully (gcc -Wall -g pearson.cpp -lstdc++ -lgsl -lgslcblas -o pearson) but when I run it here is the output:

x[0]=1.0 y[0]=1.0
x[1]=3.1 y[1]=3.2
x[2]=2.0 y[2]=1.9
x[3]=5.0 y[3]=4.9
x[4]=2.0 y[4]=2.1
Pearson correlation = 1.000000

While obviously the results should not be exactly 1, as shown with the following R code:

x <- c(1.0,3.1,2.0,5.0,2.0); y <-c(1.0,3.2,1.9,4.9,2.1)
cor(x, y, method="pearson")  # 0.99798

What am I missing?

4

1 回答 1

3

换行：

  double pearson = gsl_stats_correlation( (double*) gsl_x.vector.data, sizeof(double),
                                          (double*) gsl_y.vector.data, sizeof(double),
                                          n );

至：

  double pearson = gsl_stats_correlation( (double*) gsl_x.vector.data, 1,
                                          (double*) gsl_y.vector.data, 1,
                                          n );

或者，如果您想避免重复“幻数”1：

  const size_t stride = 1;
  double pearson = gsl_stats_correlation( (double*) gsl_x.vector.data, stride,
                                          (double*) gsl_y.vector.data, stride,
                                          n );

gsl_stats_correlation 假设double，第二个和第四个参数是“步幅”的双精度数，所以通过给它它是按字节sizeof(double)跳跃的。sizeof(double)*sizeof(double)

于 2012-01-24T00:18:53.180 回答